A Systematic Review of Integrated Risk Indicators for PET Radiopharmaceutical Production: Methodologies and Applications

Abstract

This systematic review examines the methodologies and applications of integrated risk indicators in positron emission tomography (PET) radiopharmaceutical production, focusing on occupational, technological, and environmental risks. Conducted in accordance with PRISMA 2020 guidelines and utilizing the Ryyan software 2023 for article screening, the review synthesizes findings from 70 studies published between 2020 and 2025 in English and Spanish, including articles, conference papers, and reviews. The review was registered on PROSPERO (CRD420251078221). Key disciplines contributing to risk assessment frameworks include environmental science, occupational health and safety, civil engineering, mining engineering, maritime safety, financial/economic risk, and systems engineering. Predominant risk assessment methods identified are probabilistic modeling (e.g., Monte Carlo simulations), machine learning (e.g., neural networks), multi-criteria decision-making (e.g., AHP and TOPSIS), and failure mode and effects analysis (FMEA), each offering strengths, such as uncertainty quantification and systematic hazard identification, alongside limitations like data dependency and subjectivity. The review explores how frameworks from other industries can be adapted to address PET-specific risks, such as radiation exposure to workers, equipment failure, and waste management, and how studies integrate these factors into unified risk indicators using weighted scoring, probabilistic methods, and fuzzy logic. Gaps in the literature include limited stakeholder engagement, lack of standardized frameworks, insufficient real-time monitoring, and under-represented environmental risks. Future research directions propose developing PET-specific tools, integrating AI and IoT for real-time data, establishing standardized frameworks, and expanding environmental assessments to enhance risk management in PET radiopharmaceutical production. This review highlights the interdisciplinary nature of risk assessment and the critical need for comprehensive, tailored approaches to ensure safety and sustainability in this field.

1. Introduction

Positron emission tomography (PET), a commonly used tool in nuclear medicine, has grown in popularity due to its ability to provide a three-dimensional view of bodily organs and analyze metabolic functions in tissues. The most frequently used radiopharmaceutical for PET imaging is ¹⁸F-FDG [1].

In the radiopharmaceutical industry and nuclear medicine field, the levels of radiation exposure among workers are closely tracked, specifically in terms of their effective dose and equivalent dose. This prompts the question of whether the advent of radiopharmaceuticals has an impact on occupational exposure levels [2].

As there has been a growing concern over the health effects of radiation exposure on medical workers, it has become evident that there is a pressing need for more comprehensive data on the average and long-term doses these workers are exposed to. Since medical workers are regularly exposed to low radiation doses in their daily activities, their doses are closely monitored and recorded [3].

In the production and use of PET radiopharmaceuticals (e.g., ¹⁸F-FDG), workers face varying radiation exposure risks, primarily from handling radioactive isotopes and equipment maintenance. Based on dosimetry studies and guidelines, the personnel at greatest risk are typically cyclotron engineers, who incur the highest exposures during routine servicing, target rebuilding, and handling activated components (e.g., copper mesh or Havar foils) in heterogeneous radiation fields [4].

PET radiopharmaceutical production, particularly for agents like ¹⁸F-FDG, involves unique risks stemming from the inherent properties of radioactive isotopes and the production environment. Key core characteristics include:

• Short half-life of radiopharmaceuticals: Many PET isotopes, such as ¹⁸F (half-life ~110 min), decay rapidly, necessitating just-in-time synthesis, quality control, and administration. This introduces time-sensitive risks, including production delays leading to unusable batches, increased radiation exposure during rushed handling, and logistical challenges in distribution, which can amplify technological failures (e.g., cyclotron malfunctions) and occupational hazards [5].
• Closed and aseptic nature of production processes: To maintain sterility, prevent microbial contamination, and contain ionizing radiation, PET production occurs in fully enclosed hot cells or isolators under GMP standards. This “closed system” design limits human intervention but heightens risks from equipment failures (e.g., ventilation breakdowns causing radiation leaks), process deviations (e.g., pressure imbalances), and waste management issues, where radioactive effluents must be handled without environmental release [6]. These systems also complicate real-time monitoring, potentially delaying detection of anomalies [7].
• Interplay of multiple risk dimensions: Occupational risks (e.g., chronic low-dose radiation exposure tracked via dosimetry) intersect with technological risks (e.g., synthesis module failures) and environmental risks (e.g., radionuclide transport in waste). Unlike non-radioactive pharmaceuticals, PET production requires balancing ALARA (As Low As Reasonably Achievable) principles with regulatory compliance, where even minor errors can lead to amplified consequences due to radioactivity [8].

These characteristics differentiate PET risks from other industries, demanding frameworks that incorporate dynamic, time-bound modeling and integrated monitoring to ensure safety, efficacy, and sustainability.

As mentioned in [9], effective prevention of occupational cancer requires evidence-based risk assessment and accurate estimation of exposure levels. Occupational cancer can be diagnosed through specific criteria, and interventions by occupational physicians are crucial in both prevention and early management.

Radiation protection involves three principles, justification, optimization, and dose limit. Optimization involves examining the α value, which denotes the monetary worth given to each unit of collective dose for the purpose of protecting against radiation [10]. The cost of achieving a certain level of radiation protection varies among different economic levels, making the α value closely linked to a nation’s economy [11].

Risk is a fundamental element of any major project. It exists in all projects, regardless of their scale or sector. No project can be completely free of risk. If potential threats are not properly evaluated and strategies to manage them are not developed, the project is likely to encounter failures [12]. Risk management is a structured and thorough process involving identification, analysis, planning, control, and communication of risks. Each recognized risk progresses through these stages in sequence, often simultaneously and continuously. While tracking existing risks, new ones are identified and analyzed, and mitigation strategies for one risk may introduce additional risks [13].

There are several ways to assess the risk, for example, with the perception. According to Liu et al. [14], risk perception research started in the 1960s and was initially focused on the individual level. Different research has redefined risk perception—it is influenced by social and cultural factors. Factors such as individual differences, expectations, information, risk characteristics, voluntariness, and education influence risk perception.

The evolution of occupational safety theory has shifted toward organizational responsibility, but the individual human factor remains crucial in dealing with hazards. Personality tests can aid in understanding how individuals handle activities with potential risks. Combining personality and risk perception studies can reveal key weaknesses in individual safety, supporting the development of training plans that promote safe attitudes [15].

A transdisciplinary, integrated risk assessment and risk management process is particularly beneficial to the development of policies addressing risk in complex processes. Strategies based on isolated risk assessment and management processes, guided by traditional predict then act methods for decision-making, may induce major regret if future conditions diverge from predictions [16].

This literature review aims to analyze and assess the current research on the integration of risk assessment methods and its relationship with the radiation dose in occupational exposure. By examining existing literature, it is possible to gain a deeper understanding of how these methods have been used and their effectiveness in accurately estimating the radiation dose. This is crucial in ensuring the safety and well-being of individuals working in occupations with potential exposure to radiation. The following five research questions were formulated:

(Q1) What are the primary disciplines that have developed integrated risk indicators applicable to PET radiopharmaceutical production, and how do their methodologies differ?

(Q2) Which risk assessment methods (e.g., probabilistic modeling, machine learning, or multi-criteria decision-making) are most used to evaluate risks and what are their strengths and limitations?

(Q3) How can existing integrated risk indicator frameworks from other industries (e.g., environmental science, nuclear safety, or maritime engineering) be adapted to address the specific occupational, technological, and environmental risks in PET radiopharmaceutical production?

(Q4) How do current studies quantify and integrate diverse risk factors (e.g., radiation exposure, equipment failure, and waste management) into a unified risk indicator for PET radiopharmaceutical production?

(Q5) What are the gaps in the literature regarding the application of integrated risk indicators in PET radiopharmaceutical production, and what future research directions can address these deficiencies?

2. Material and Methods

This systematic review adheres to established guidelines to ensure methodological rigor and transparency. The Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) framework was employed as the cornerstone of our methodology. PRISMA is a widely recognized set of evidence-based guidelines designed to improve the reporting of systematic reviews, promoting completeness, clarity, and reproducibility by outlining steps for literature identification, screening, eligibility, and inclusion [17]. In this review, PRISMA guided our search strategy, flow of study selection (as depicted in Figure 1), and data synthesis.

Additionally, the Population, Intervention, Comparator, Outcome (PICO) framework informed the development of our research questions and inclusion criteria. PICO is a structured approach commonly used in evidence-based practice to define key elements of a review question: Population (e.g., industries or contexts involving risk assessment, such as PET radiopharmaceutical production), Intervention (e.g., integrated risk indicator methodologies), Comparator (e.g., conventional vs. advanced risk assessment methods), and Outcome (e.g., improvements in risk management, safety, and sustainability).

To facilitate efficient management of retrieved records, the Rayyan software 2023 (Rayyan Systems Inc., Cambridge, MA, USA) was employed. This web-based tool enabled the initial upload of citations from multiple databases, automated detection and removal of duplicates, and collaborative screening by the review team. Titles and abstracts were independently assessed for relevance using predefined inclusion and exclusion criteria, with conflicts resolved through consensus discussions. Full-text articles were then evaluated in Rayyan for final eligibility, ensuring a systematic and transparent filtering process that minimized bias and enhanced reproducibility [18].

This systematic review was prospectively registered in the International Prospective Register of Systematic Reviews (PROSPERO) database to ensure transparency and adherence to predefined methods. The registration ID is CRD420251078221, and the title is “A Systematic Review of Integrated Risk Indicators for PET Radiopharmaceutical Production: Methodologies and Applications”.

Data were documented according to a standardized protocol, where objectives and inclusion criteria were specified in detail.

Eligibility Criteria:

The following eligibility criteria were used for selecting studies to be included in this systematic review:

• Language: English and Spanish.
• Publication Date: 2020–2025.
• Document Type: articles, conference papers, and reviews (Scopus allows conference papers, which may be relevant for technological risks).
• Subject Area: limit to relevant fields (e.g., Environmental Science, Medicine, Engineering, and Social Sciences) if needed.

Research Question (using the PICO framework):

• Population: workers, communities, or systems exposed to risks.
• Intervention/Exposure: assessment of subjective and objective risks, environmental factors, and occupational exposures.
• Comparison: studies with or without integral risk indicators.
• Outcome: development or evaluation of an integral risk indicator.

2.1. Search Strategy

Subjective risk (risk perception):

• Keywords: “risk perception”, “perceived risk”, “subjective risk”, “risk attitude”, “risk awareness”, and “psychological risk”.

Objective risk (technological risk):

• Keywords: “technological risk”, “objective risk”, “technical risk”, “system failure”, “equipment risk”, “cybersecurity risk”, and “industrial risk”.

Environmental indicators:

• Keywords: “environmental risk”, “environmental indicators”, “pollution”, “climate risk”, “ecological risk”, and “sustainability indicators”.

Occupational exposure indicators:

• Keywords: “occupational exposure”, “workplace risk”, “occupational hazard”, “chemical exposure”, “ergonomic risk”, and “health and safety indicators”.

Integral risk indicator:

• Keywords: “integral risk indicator”, “composite risk index”, “risk assessment framework”, “multidimensional risk”, and “integrated risk model”.

2.2. Inclusion Criteria

• Studies addressing at least two of the four components (subjective risk, objective risk, environmental indicators, and occupational exposure).
• Studies discussing risk assessment or integrated risk models.

2.3. Exclusion Criteria

• Studies focusing on only one component without integration.
• Non-peer-reviewed sources or abstracts without full text.

3. Results and Discussion

This analysis addresses the development and application of integrated risk indicators for PET radiopharmaceutical production, drawing on a systematic review of 70 studies from 2020 to 2025, screened from 6017 articles across Scopus, IEEE, INIS, and PubMed using the PRISMA method, as documented in Figure 1. Selected papers from the systematic review to research the use of the risk assessment methods and radiation applications are shown in

Table 1. Selected papers from the systematic review to research the use of the risk assessment methods and radiation applications.

No.	Author (Year) [Reference]	Subjective Risk Component	Objective Risk Component	Environmental Risk Component	Occupational Risk Component	Integrated Risk Indicator
1	Anwar et al. (2020) [19]	None: No risk perception or subjective judgments (pp. 1–19)	Full: PBEE with fiber-based modeling, IDA, Monte Carlo simulations, fragility functions (pp. 2–8, 14)	Full: Equivalent CO2 emissions from repair materials (pp. 3, 10–11)	None: No occupational risk focus (pp. 1–19)	Yes: Seismic resilience impact (SI), resilience (RS), integrating economic, social, environmental factors (pp. 3–4, 14)
2	Padgett et al. (2020) [20]	None: No risk perception focus (pp. 1–23)	Full: Probabilistic LCS-A, surrogate models (PRSM, logistic regression), Monte Carlo, LHS (pp. 2–10, 14–15)	Full: Embodied energy for environmental impact (pp. 2, 14–16)	None: No occupational risk focus (pp. 1–23)	Yes: Life-cycle sustainability (LCS), interaction effects, integrating phase contributions (pp. 2–3, 16–17)
3	Bajenescu (2020) [21]	Partial: Risk appetite as managerial willingness to accept risks (pp. 6–7)	Full: PRA, FMEA, Fault Tree Analysis, ALT, probabilistic modeling (pp. 7–12, 15)	Full: Environmental pollution, disaster risk reduction (pp. 5–6, 12–13)	Partial: Safety risks to personnel (pp. 5–6)	Yes: Risk triplets, performability, integrating scenarios, likelihood, consequences (pp. 7, 16–17)
4	Ghoushchi et al. (2020) [22]	Full: Expert linguistic judgments via Z-number theory (pp. 2–3, 9–12)	Full: Z-SWARA, Z-MOORA, FMEA, triangular fuzzy numbers (pp. 3–12, 18–20)	None: No environmental focus (pp. 1–28)	Full: Warehousing risks (e.g., pallet falling, equipment collisions) (pp. 18–20)	Yes: Z-MOORA scores, integrating S, O, D, C, T (pp. 11–12, 19–20)
5	Demich et al. (2020) [23]	Partial: Workers’ hazard perceptions via near-miss reports (pp. 8–9)	Full: 5 × 5 RA matrix, nonparametric median test, Hierarchy of Controls (pp. 5–10)	Partial: Workplace environmental hazards (e.g., housekeeping) (pp. 4, 6)	Full: Mining occupational risks (e.g., equipment failure, PPE) (pp. 1–10)	Yes: Risk rankings (low, moderate, high, critical), integrating probability, consequence, corrective actions (pp. 6–7)
6	Meacham et al. (2021) [24]	Partial: Public/expert risk perception differences (pp. 2, 92–99)	Full: STBRSAM with 86 diagnostic factors, 0–4 rating (pp. 8–10)	Partial: Fire-related environmental impacts (pp. 5, 143–145)	Partial: Operator competency, safety management (pp. 11, 314–321)	Yes: STBRSAM composite scores (p. 10, Figure 4)
7	Hardy et al. (2020) [25]	Full: Discursive risk translation based on stakeholder perceptions (pp. 2–4, 6–10)	Partial: Qualitative coding (descriptive, analytical, pattern) of interviews, texts (pp. 6, 8–9)	Full: BPA’s endocrine-disrupting environmental risks (pp. 5, 7, 10)	None: No occupational focus (pp. 1–32)	Yes: Ecology of risks, integrating translated risks (professional, regulatory, reputational, operational) (pp. 2, 24)
8	Montaño et al. (2020) [26]	Partial: Staff operational observations (pp. 1–2)	Full: HAZOP methodology, risk matrix, P&ID analysis (pp. 1–3)	Full: Environmental damage from processes (p. 3)	Full: Personnel injuries (e.g., burns, fractures) (p. 3)	Yes: Risk matrix ranking (Very High, High, Medium, Low, None), integrating severity, likelihood (pp. 3–4)
9	Ancione et al. (2021) [27]	None: No risk perception focus (pp. 1–15)	Full: Bayesian Network, GIS, Counting Learning, k-out cross-validation (pp. 3–9)	Full: Chemical releases into air, soil, water (pp. 1–2, 12–13)	Partial: Worker presence in impact areas (pp. 5–6)	Yes: Na-Tech Risk Index (R1–R4), integrating hazard, vulnerability, release, damage (pp. 5–11)
10	Bilgin (2021) [28]	Full: Expert linguistic judgments via fuzzy logic (pp. 4–5, 8)	Full: Entropy-weighted fuzzy FMEA, triangular fuzzy numbers, COA defuzzification (pp. 4–6, 8–9)	None: No environmental focus (pp. 1–12)	Full: Occupational risks (e.g., hand injuries, slipping) (pp. 7–9)	Yes: Entropy-weighted fuzzy RPNs, integrating severity, occurrence, detection (pp. 4, 9)
11	Brocal Fernandez et al. (2021) [29]	None	Full: Literature-based classification, dynamic risk analysis, Bayesian Nets, PDVA cycle	Partial: ISO 14001, nanomaterial concentration	Full: Occupational risk indicators, nanomaterials	Yes: Leading indicator classification, specific metrics (e.g., percentage of risks within specifications), conceptual scheme, synthesizing process integrity, occupational, management risks
12	Fabis-Domagala et al. (2021) [30]	Partial: Expert subjectivity in detection estimation (pp. 2–3, 7)	Full: Modified FMEA, criticality number, failure rates, normalization (pp. 3–7)	None: No environmental focus (pp. 1–16)	Partial: Safety implications of failures (pp. 3, 9–11)	Yes: Criticality number (Cr), integrating severity, failure predictor, detection (pp. 3–11)
13	Ho et al. (2021) [31]	None: No risk perception focus (pp. 1–24)	Full: Mechanistic (RWQM1, ASM1) and fuzzy rule-based models, Monte Carlo, GLUE, hill-climbing (pp. 8–11, 15–19)	Full: GHG (CO2, CH4, N2O) emissions in urban rivers (pp. 3–4, 15–16)	None: No occupational focus (pp. 1–24)	Yes: Qualitative risk levels (low, moderate, high), integrating DO, WQI, flow velocity (pp. 11–15)
14	Kaylaniet al. (2021) [32]	Full: Expert opinions via questionnaires (pp. 5–7)	Full: Hybrid FMEA-AHP, RPN, pairwise comparisons, normalized eigenvectors (pp. 5–9)	Partial: Indirect via environmental benefits (pp. 1, 3)	Partial: Safety implications of failures (pp. 2–3)	Yes: RPN with AHP weights, integrating severity, occurrence, detection (pp. 6–12)
15	Nobanne et al. (2021) [33]	Full: Subjective risk perception analysis (pp. 1–2, 10–12)	Full: Bibliometric analysis, VOS viewer, descriptive statistics (pp. 1–6, 13–15)	Partial: Indirect via cited works (e.g., flood risk) (pp. 7–12)	None: No occupational risk focus (pp. 1–20)	Yes: Clustered keyword streams integrating subjective/objective risk themes (pp. 10–12)
16	Ngo et al. (2021) [34]	Full: Stakeholder risk perceptions in co-design (pp. 1, 36–42; pp. 12, 374–380)	Full: 1D/2D hydraulic modeling, depth-damage curves (pp. 5, 151–169)	Full: Flood hazards, climate change scenarios (pp. 6–8, 198–245)	None: No workplace hazard focus (pp. 1–16)	Yes: Flood water levels, inundation/damage maps, damages (pp. 6–8, Figure 2, Table 1)
17	Patel et al. (2023) [12]	Partial: Questionnaire surveys, expert opinions	Full: RII, brainstorming, Delphi technique, interviews, case studies	Full: Weather, natural calamities, site access	Full: Accidents due to poor safety procedures, labor accidents	Yes: RII rankings (e.g., financial failure, RII = 0.791), risk impact assessments, integrating technical, environmental, financial, occupational risks
18	Ying et al. (2021) [35]	Partial: Likert 5-scale questionnaire for indicator importance (pp. 6, 495–504)	Full: Fuzzy evaluation, catastrophe progression, factor analysis (pp. 6–8)	Full: Pipeline accident pollution risks, disaster factors (pp. 1, 4)	Partial: Personnel safety quality, health status (pp. 5, 304–316)	Yes: Individual risk value (IR) via catastrophe progression (pp. 8–9)
19	Wang (2021) [36]	Full: Expert assessments, questionnaire surveys (pp. 2, 5–6)	Full: Multi-level fuzzy assessment, direct weight method, fuzzy matrices (pp. 3–7)	Partial: Public utilities failure (p. 5)	Partial: Labor disputes, talent outflow (p. 5)	Yes: Fuzzy assessment scores (B, E), integrating strategy, procurement, manufacturing, distribution risks (pp. 6–7)
20	Alizadeh et al. (2022) [37]	Partial: Expert opinions for fuzzy functions (pp. 4–5)	Full: Fuzzy FMEA, MATLAB (R2021b) fuzzy inference system, RPN calculation (pp. 4–7)	Full: Effluent discharge, contaminant release (pp. 2, 11)	Full: Worker safety hazards from equipment failures (pp. 4, 11)	Yes: Fuzzy RPN, integrating severity, occurrence, detection (pp. 7–11)
21	Bortey et al. (2022) [38]	Full: Risk perception, safety climate, culture (pp. 4, 6, 11)	Full: Scientometric analysis, VOS Viewer, machine learning (e.g., Bayesian networks, SVM) (pp. 2–12)	Partial: Dynamic construction environment (pp. 1–3)	Full: Highway worker safety, human errors, injuries (pp. 1, 4–6)	Yes: Resilient predictive safety risk score, integrating human, technical, situational risks (pp. 11–12)
22	Costola et al. (2022) [39]	Full: Public concern via Google SVIs (pp. 2, 5)	Full: Diebold-Yilmaz FEVD, VAR, Fama-MacBeth regressions (pp. 2–3, 21)	Partial: WEF environmental risks (pp. 2, 5)	None: No occupational focus (pp. 1–33)	Yes: GRAI, net directional spillover indices, risk premium estimates, synthesizing economic, environmental, geopolitical, societal, technological risks (pp. 19–23)
23	Liu et al. (2022) [40]	Partial: Expert scoring by professionals (pp. 5–6)	Full: System dynamics, Vensim, five-dimensional risk measurement, entropy weights (pp. 3–11)	Full: Environmental protection costs, pollution fines (pp. 5, 7)	None: No occupational focus (pp. 1–15)	Yes: Risk entropy, integrating political, financial, market, environmental risks (pp. 5–11)
24	Lofti et al. (2022) [41]	None: No risk perception focus (pp. 2–8)	Full: Hybrid robust stochastic programming, CVaR, GAMS-CPLEX, linearization (pp. 4–8)	Full: Pollution optimization (e.g., CO2 emissions) (pp. 2, 9–15)	None: No occupational focus (pp. 1–18)	Yes: Weighted objective function with CVaR, integrating time, cost, quality, energy, environment (pp. 7–15)
25	Petronijevic et al. (2022) [42]	Partial: Expert judgments in risk factor identification and fuzzy cognitive map creation (pp. 7220, 7225)	Full: Probabilistic modeling, fuzzy cognitive maps, agent-based simulation for risk assessment (pp. 7215, 7220–7225)	Partial: Environmental risks indirectly addressed through resource availability and process impacts (pp. 7216, 7225)	Partial: Occupational risks considered through resource and task behavior modeling (pp. 7216, 7225)	Yes: Simulation outputs including cost, time, and value risks, integrating diverse risk factors via fuzzy cognitive maps and Bayesian networks (pp. 7220, 7225–7226)
26	Renn et al. (2022) [43]	Full: Risk perception, social constructs (pp. 11–13)	Full: Complexity modeling, scenario construction, empirical analysis (pp. 6–9, 13–14)	Full: Climate change, biodiversity loss (pp. 8–9)	None: No occupational focus (pp. 1–19)	Yes: Systemic risk indicators in scenarios, integrating complexity, uncertainty, ambiguity (pp. 3–4, 13–14)
27	Santillan-Saldivar et al. (2022) [44]	None: No risk perception focus (pp. 1–9)	Full: GeoPolRisk, GeoPolEndpoint, openLCA, CML 2001, ReCiPe (pp. 4–6)	Full: Environmental impacts via CML 2001, ReCiPe (pp. 6–8)	None: No occupational focus (pp. 1–13)	Yes: GeoPolRisk (midpoint), GeoPolEndpoint (endpoint), integrating supply risk, mass flows, price elasticity (pp. 6–9)
28	Wang et al. (2022) [45]	Partial: Expert scoring for qualitative factors (pp. 5–6)	Full: Fuzzy set theory, AHP, membership functions, fuzzy transformation (pp. 3–7)	Full: Mine ventilation, dust, gas, water hazards (p. 3)	Full: Staff operations, training, equipment hazards (p. 3)	Yes: Fuzzy comprehensive evaluation score, integrating 17 safety factors (p. 7)
29	Zdzislawa (2022) [46]	Partial: Expert assessments, Delphi method (pp. 2, 7–11)	Full: EFMEA, RPN calculations, VIKOR, DEMATEL, ANP, AHP (pp. 2–8)	Full: Air, water, soil pollution, waste production (pp. 2–10)	Partial: Health/safety risks in some studies (pp. 7, 11)	Yes: RPN (S × O × D), integrating severity, occurrence, detection via multi-criteria methods (pp. 2, 7–11)
30	Yasbayir et al. (2022) [47]	Partial: Expert evaluations for S, O, D (pp. 3–8)	Full: MGT-GFMEA, GRA, mass gravity equations, weighted RPN calculations (pp. 5–12)	None: No environmental focus (p. 8)	Partial: Worker-related failure modes (e.g., welding defects) (p. 8)	Yes: Weighted RPNs, integrating S, O, D via mass gravity weights (pp. 11–12)
31	Zhang et al. (2022) [48]	Partial: Expert interviews, questionnaires (pp. 6, 11)	Full: PP-PSO, improved SPA, five-element connection numbers, partial derivatives (pp. 7–10)	Full: Geological, hydrological conditions (pp. 5–6)	Partial: Human-related risks (pp. 5–6)	Yes: Weighted five-element connection number, integrating human, material, equipment, method, environment (pp. 9–13)
32	Zhu et al. (2023) [49]	None: No risk perception focus (pp. 1–18)	Full: System dynamics, Rough Set, ANP, FCE, Vensim PLE (pp. 5–10)	Full: Mine environmental hazards (pp. 5–6, 12)	Full: Worker safety, training risks (pp. 5–6, 12–15)	Yes: Safety risk management level, integrating subsystem impacts (pp. 9–12)
33	Cai et al. (2023) [50]	Partial: Expert interviews for validation (pp. 16–18)	Full: Two-mode network, WBS-RBS, OBS-WBS, centrality analysis (pp. 6–11)	Partial: Geological, ventilation hazards (pp. 11–12)	Full: Worker safety risks in tunnels (pp. 1–2, 12, 18–19)	Yes: Weighted/unweighted degree centrality, integrating risk-sharing relationships (pp. 10–15)
34	Chu et al. (2023) [51]	None: No risk perception focus (pp. 1–26)	Full: NRAP-IAM-CS, RROM-Gen, ECLIPSE simulations, Monte Carlo, ROMs (pp. 2–5, 9–12)	Full: CO2, brine leakage to groundwater (pp. 9–20)	None: No occupational focus (pp. 1–26)	Yes: Cumulative CO2, brine leakage percentages, integrating well permeability, reservoir pressure, CO2 saturation (pp. 12–20)
35	Dai et al. (2023) [52]	None: No risk perception focus (pp. 1–24)	Full: InSAR (Stacking, SBAS), neural networks, certainty factors, matrix combination (pp. 4–7, 16–17)	Full: Landslide impacts on natural environments (pp. 10–12, 18–19)	None: No occupational focus (pp. 1–24)	Yes: Hazard level (H), integrating susceptibility, temporal/event probabilities, deformation rates (pp. 6–7, 18)
36	Duda et al. (2023) [53]	Partial: Employee experience in hazard assessment (pp. 9–10)	Full: FMEA, RPN calculations (pp. 2–5)	Partial: Methane, rock burst impacts (pp. 2, 4–6)	Full: Worker injuries, fatalities (pp. 4–9)	Yes: RPN (E × O × D), integrating severity, occurrence, detection (pp. 4–9)
37	Gao et al. (2023) [54]	Partial: Household risk perception for preparedness (pp. 2, 10)	Full: DFNN, LIME, Adam optimizer, softmax activation (pp. 3–6)	Full: Hurricane wind, storm surge hazards (pp. 3–4, 10)	None: No occupational focus (pp. 1–13)	Yes: Hurricane risk levels (0–5 wind, 0–6 surge), integrating building, meteorological, hydrological features (pp. 5–9)
38	Tan et al. (2023) [55]	Partial: Expert opinions via questionnaires (pp. 4–6)	Full: GBWM, GIS, MCDM, SAW method (pp. 3–8)	Full: Chemical spill risks to coastal ecosystems (pp. 1–2, 7–9)	None: No occupational focus (pp. 1–13)	Yes: Hazard score (H), integrating chemical criteria, weights via GBWM, GIS mapping (pp. 7–9)
39	Özdemir (2023) [56]	Partial: Expert interviews, assessments (pp. 4, 6)	Full: FMEA, RPN calculations (pp. 4–6)	Partial: Terrain, sun exposure, wildlife risks (pp. 7–9)	Full: Bee stings, ergonomic, chemical risks (pp. 3–11)	Yes: RPN (E × P × D), integrating severity, probability, detection (pp. 6–9)
40	Penserini et al. (2023) [57]	None: No risk perception focus (pp. 1–10)	Full: QCRA, Monte Carlo, MLE_LC, R software (R-4.2.2) for Windows (NADA package) (pp. 5–7)	Full: Alkylphenol contamination across water, soil, crops (pp. 2–3)	None: No occupational focus (pp. 1–10)	Yes: Benchmark Quotient (BQ), integrating exposure dose, HBGV, uncertainty via Monte Carlo (pp. 6–8)
41	Tajasosi et al. (2023) [58]	Partial: Expert assessments for risk weighting (pp. 8–9)	Full: FTA, semi-quantitative risk index, Life-365 simulations (pp. 6–10)	Full: Material emissions, CO2 footprint (pp. 8–10)	Partial: Human error in production (p. 7)	Yes: Overall Risk Index (ORI), integrating economic, technical, environmental risks (pp. 9–14)
42	Toktaş et al. (2023) [59]	Full: Expert evaluations via DEMATEL (pp. 3, 27–29)	Full: KEMIRA-M, DEMATEL, weighted normalized vectors (pp. 3–4, 24–26)	Partial: Workplace environmental hazards (e.g., thermal conditions) (pp. 24, 29–30)	Full: Work accidents, occupational diseases (pp. 1–2, 24, 27–30)	Yes: Weighted vectors, danger source weights, integrating risk criteria, danger sources, measures (pp. 25–26)
43	Wang et al. (2023) [60]	Full: Expert opinions, user behavior via Delphi, questionnaires (pp. 7, 10–11)	Full: STPA-FMEA, RPN calculation, fuzzy evaluation (pp. 4–5, 9–12)	Partial: Network, physical, social environmental risks (pp. 6–7, 13–14)	None: No occupational focus (pp. 1–19)	Yes: Risk Priority Number (RPN), integrating severity, occurrence, detectability, user, environmental factors (pp. 9–12)
44	Wattanayon et al. (2023) [61]	None: No risk perception focus (pp. 1–16)	Full: WBE, HPLC-MS/MS, RQ method, correction factors (pp. 2, 8–11)	Full: AF contamination in wastewater, rivers (pp. 2, 8–11)	None: No occupational focus (pp. 1–16)	Yes: Risk Quotient (RQ), PNDLs, integrating environmental concentrations, PNECs, exposure estimates (pp. 11–14)
45	Abuhussain (2024) [62]	Full: Expert evaluations via Likert-scale surveys (pp. 5–6, 13)	Full: Fuzzy TOPSIS, EANN, TFNs, loss function analysis (pp. 7–10, 15)	Full: Weather, geological conditions, material impacts (pp. 6–7, 13)	Full: Accidents, human errors, safety equipment (pp. 6–7, 13)	Yes: Closeness coefficient, EANN rankings (pp. 8–9, 16)
46	Chen et al. (2024) [63]	None: No risk perception data (pp. 1–17)	Full: BP neural network, SVM, wearable sensor data (pp. 7–14)	Partial: Simulated high-altitude conditions via VR (pp. 5–6)	Full: Worker fatigue, health risks (pp. 8–12)	Yes: Binary risk classification (safe vs. dangerous), integrating physiological and demographic factors (pp. 13–14)
47	Chang et al. (2024) [64]	Partial: Expert feedback, interviews for risk identification (pp. 4, 12)	Full: k-means clustering, MCDA with AHP, Monte Carlo, AI models (pp. 14, 18, 20–21)	Full: Climate hazards (e.g., floods, precipitation), CIVS (pp. 1, 20–22)	None: No focus on worker safety (pp. 1–28)	Yes: CIVS, integrating environmental factors via AHP and clustering (pp. 20–22)
48	Jin et al. (2024) [65]	Full: Linguistic evaluations, PIS, expert interviews (pp. 6–10, 13–14)	Full: Group FMEA, PIS, CRP, distributed linguistic functions, linear programming (pp. 7–12, 19)	Full: Weather, sea conditions, LNG leak impacts (pp. 13, 27)	Full: Mariner skills, training, fitness risks (pp. 13, 27)	Yes: Composite risk scores (e.g., MINLU: 2.780), integrating occurrence, severity, detectability via weighted FMs (pp. 20–22)
49	Kumi et al. (2024) [66]	Full: Surveys, interviews, capturing risk perceptions (p. 10)	Full: Statistical analysis, mathematical modeling, simulation, AI (pp. 8–9)	Full: Site conditions, environmental data from sensors (pp. 9–10)	Full: Worker safety, behavior, injury risks (pp. 11–12)	Yes: Risk scores, fatality rates, probability metrics via multiple methods (pp. 11–12)
50	Bohrey et al. (2024) [67]	Full: Expert linguistic judgments by AMEs (pp. 4–5)	Full: Hybrid FMEA-FAHP-FTOPSIS, HFACS-ME (pp. 2–7)	Partial: Workplace hazards (e.g., lighting, noise) (pp. 3, 9)	Full: Maintenance errors, injuries (pp. 3–9)	Yes: Closeness Coefficient Index (C_C), integrating severity, occurrence, detection (pp. 6–7)
51	Cid-Escobar et al. (2024) [68]	Partial: Household surveys for socio-economic factors (pp. 5, 13–14)	Full: Transient groundwater modeling, GIS, MFA, FactorMineR (pp. 3–5, 12–13)	Full: Groundwater scarcity, quality degradation (pp. 1–2, 6, 12–14)	None: No occupational focus (pp. 1–17)	Yes: Composite risk index (A × S × P), integrating availability, sustainability, proximity (pp. 4–5, 12–13)
52	Kang et al. (2024) [69]	Partial: Expert and staff questionnaire scores (pp. 8–9, 15)	Full: GA-BP, PSO-BP, WOA-BP, RF, SVR, 5-fold cross-validation (pp. 9–15)	Partial: External changes, catastrophic events (pp. 5, 16)	Full: Employee negligence, errors, training deficits (pp. 5, 16–17)	Yes: Risk level (Y), integrating 17 indicators via GA-BP (pp. 8–9, 15–16)
53	Abbasi et al. (2024) [70]	Partial: Safety team judgments for event identification (pp. 1, 4–5)	Full: FFTA, FBN, Bow-tie, PHAST/SAFETI, GIS zoning (pp. 4–6, 10–12)	Full: Explosion impacts on urban areas (pp. 1, 12–15)	Partial: Inadequate supervision, safety plan deficits (p. 15)	Yes: Individual/societal risk levels (e.g., 10 × 10−5), integrating 33 events via FBN (pp. 12–15)
54	Kumar et al. (2024) [71]	Full: Worker surveys via WRD questionnaire (pp. 2–3, 12–13)	Full: EFA, Pearson correlation, multiple linear regression (pp. 3–5)	Partial: Workplace conditions (e.g., lighting, ventilation) (pp. 5–6, 12)	Full: Injuries, job stress, safety training (pp. 1–2, 5–9)	Yes: Normalized priority weights (ω_i), integrating occupational risks via expert ratings (pp. 4–8)
55	Nashira al. (2024) [72]	Partial: Semi-structured interviews with operators, management, safety committee (pp. 3–4)	Full: FMEA, RPN calculations (1–125 scale), risk register with 34 risks (pp. 4–8)	Full: Effluent quality degradation, untreated wastewater discharge (pp. 6–9)	Full: Operator exposure to bioaerosols, inadequate PPE, non-compliance with SOPs (pp. 1–2, 8–9)	Yes: RPNs (acceptable: 1–15, undesirable: 16–30, unacceptable: >30, e.g., R08: 45, R21: 20), risk mapping, synthesizing human, financial, environmental risks (pp. 4–10)
56	Tepparit et al. (2024) [73]	Partial: Expert brainstorming, user satisfaction assessments (pp. 3, 8)	Full: Python-based risk assessment program, check sheets, JSA, COSO-ERM, Lean ECRS (pp. 5–9)	Partial: Wastewater pollution, community impacts (pp. 2, 4)	Full: Safety hazards (e.g., electrical shock, pinch points), JSA analysis (pp. 4, 8)	Yes: Risk levels (low: 1–2, moderate: 3–4), integrating likelihood and impact via check sheets and JSA (pp. 5–8)
57	Qi et al. (2024) [74]	Full: Driver surveys (5274 responses) via Likert scale (pp. 2–4)	Full: BP neural network, 8-fold cross-validation, L2 regularization (pp. 5–15)	Full: Road congestion, weather conditions (pp. 8–9)	Full: Driver errors, fatigue, speeding risks (pp. 7–8, 16)	Yes: Risk level (0–1), integrating 10 indicators via BP (pp. 15–16)
58	Wang et al. (2024) [75]	None: No risk perception data (pp. 1–15)	Full: Upscaling, streamline models, RSA, Monte Carlo, PSUADE (pp. 2–5)	Full: Radionuclide transport risks to biosphere (pp. 1–2, 12–13)	None: No worker safety focus (pp. 1–15)	Yes: Release dose, breakthrough time, peak dose, peak time (pp. 5, 12–13)
59	Weng et al. (2024) [76]	Full: Expert interviews, 314 survey responses (pp. 2, 18–19)	Full: PCA, CRITIC-EWM, SEM validation (pp. 9–24)	Partial: Environmental pollution, adaptability (pp. 6, 16–17)	None: No occupational focus (pp. 1–27)	Yes: Weighted indicator system, integrating 21 indicators via CRITIC-EWM (pp. 16–24)
60	Yazo-Cabuya et al. (2024) [77]	Full: Expert surveys for risk prioritization (pp. 9–10)	Full: DEMATEL, AHP, sensitivity analysis (pp. 10–20)	Full: Carbon emissions, water depletion (pp. 7–8, 21)	Partial: Safety, health at work (pp. 7–8, 21)	Yes: DEMATEL weightings, AHP priority scores, integrating risk typologies, sub-risks (pp. 12–21)
61	Yu et al. (2024) [78]	Full: Expert surveys via Literature-Delphi (pp. 3, 7–8)	Full: AHP, entropy, fuzzy comprehensive evaluation (pp. 3–7)	Full: Hydrometeorological, natural disasters, interference (pp. 5, 7–8)	Partial: Technical personnel, operator knowledge (pp. 5, 7)	Yes: Risk score (T), integrating 27 indicators via fuzzy matrices (pp. 7–8, 10)
62	Eid et al. (2025) [79]	None: No risk perception data (pp. 1–22)	Full: SOM, PCA, cluster analysis, Monte Carlo simulation (pp. 4, 20)	Full: Salinization, heavy metal contamination, industrial indices (pp. 3–4, 20)	None: No focus on worker safety (pp. 1–22)	Yes: CWQI, HI, MPI, NCI, integrating physicochemical and health risks (pp. 3–4, 20)
63	Erdem et al. (2025) [80]	Partial: Expert judgments for fuzzy AHP (pp. 3, 8)	Full: Fuzzy AHP, TOPSIS, TFNs (pp. 8–10)	Full: Emissions, pollution, renewable energy (pp. 11–12, 21)	Full: Safety risks, equipment failures (pp. 11, 19)	Yes: OPS via TOPSIS, integrating ten criteria (pp. 10, 18–19)
64	Lai et al. (2025) [81]	Full: Expert surveys for risk scoring (pp. 10, 32)	Full: System dynamics, G1, entropy methods (pp. 6–15)	Full: Environmental regulations, ecological impacts (pp. 8, 29–30)	Partial: Site safety risks (pp. 8, 29)	Yes: Composite subsystem risk scores via SD (pp. 14–15)
65	Jing et al. (2025) [82]	None: No explicit reliance on subjective risk perceptions or expert judgments (pp. 3–4)	Full: HFACS for causal factor classification, Apriori algorithm for association rule mining, social network analysis with Gephi (v 0.10.1) for visualization, network centrality, and core-periphery structure analysis (pp. 3–7)	Full: Environmental risks addressed through factors like methane accumulation and ventilation system inefficiencies (pp. 5–7)	Full: Occupational risks identified, including unsafe human behaviors and supervision failures (pp. 5–9)	Yes: Core-periphery structure with 21 core and 32 peripheral risk factors, integrating human, environmental, and organizational risks via association rules and network metrics (pp. 7–10)
66	Mohsin et al. (2025) [83]	Full: Stakeholder risk perception via survey (pp. 1, 7)	Full: SEM, CFA, SPSS (V. 30.0.0), AMOS (pp. 1, 5–9)	Full: Climate change, pollution, overexploitation (pp. 3, 9)	Partial: Fishermen’s economic, environmental risks (pp. 7, 11)	Yes: Composite risk impact score via SEM, integrating five risk types (pp. 9–11)
67	Najar et al. (2025) [84]	None: No risk perception or subjective judgments (pp. 1–10)	Full: HotSpot Code (v3.1.2), ANN for accident identification, LSTM for dose prediction, Gaussian plume modeling (pp. 3–8)	Full: Radionuclide deposition, soil roughness (3–300 cm), atmospheric dispersion (pp. 2, 5–9)	Partial: AI-enhanced operator reliability to reduce human error (pp. 3, 6)	Yes: TEDE, GSD, integrating radionuclide concentrations, soil roughness, distance, with LSTM forecasting negligible doses (pp. 5–9)
68	Wang et al. (2025) [85]	Partial: Expert judgments in AHP hierarchy setup (pp. 3, 10)	Full: GRA-improved AHP, BN with T-S fuzzy fault tree, game theory combined weighting, implemented in Genie2.3 (pp. 3–10)	Full: Soil moisture, chloride, sea water mineralization, sand content (pp. 1, 10–11)	None: No occupational focus (pp. 1–14)	Yes: Combined weights via game theory, integrating AHP and BN weights (e.g., C3, C4: 0.0575) (pp. 10, 13)
69	Yang et al. (2025) [86]	Partial: Expert knowledge and officer sensitivity in risk functions (pp. 3, 20)	Full: Spatiotemporal risk modeling with AIS data, asymmetric Gaussian-based risk functions, adaptive risk fusion (pp. 1, 14–20)	Full: Waterway width, channel boundaries, non-navigable areas (pp. 18–20)	Partial: Officer decision-making in collision avoidance (pp. 16–17, 20)	Yes: Total risk value via multi-source fusion, integrating static and dynamic risks (pp. 18–20)
70	Zhou et al. (2025) [87]	None: No risk perception data (pp. 1–22)	Full: CRITIC-Cloud Model, SPA, Python 3.9 (pp. 2–3, 7–11)	Full: Ground subsidence, building settlement (pp. 4–5, 17)	Partial: Structural stability risks	Yes: Composite risk level via μ and Kp (pp. 11, 16–17)

.

(Q1) What are the primary disciplines that have developed integrated risk indicators applicable to PET radiopharmaceutical production, and how do their methodologies differ?

Integrated risk indicators are methodologies that synthesize multiple risk dimensions, subjective (e.g., risk perception), objective (technical), environmental, and occupational, into a unified risk assessment framework. The following disciplines have been identified as primary in developing integrated risk indicators, based on the context and focus of the studies:

• Environmental Science/Engineering: Environmental science and engineering significantly impact PET radiopharmaceutical production by addressing sustainability, waste management, contamination prevention, and regulatory compliance. Key influences include:
  • Waste management and radionuclide transport: These disciplines develop models to assess and mitigate environmental risks from radioactive effluents, such as groundwater contamination from short-lived isotopes like ¹⁸F (half-life ~110 min) [88]. For instance, radionuclide transport assessments minimize leakage during production and disposal, ensuring compliance with GMP and ALARA [89].
  • Eco-friendly processes and decontamination: Innovations like graphene oxide-based purification or solid-phase materials reduce chemical waste and enable greener labeling/synthesis methods, lowering the ecological footprint of cyclotron operations and imaging agents [90].
  • Energy consumption and sustainability: Assessments highlight high energy use in PET/CT systems, promoting strategies like optimized production schedules or renewable integrations to cut CO₂ emissions and resource depletion [91]
  • Risk reduction and monitoring: Integration of IoT sensors and AI for real-time detection of leaks or anomalies enhances environmental safety, preventing bioaccumulation of contrast agents [91].

Environmental science/engineering studies often integrate technical and environmental components, with less emphasis on subjective or occupational risks, but directly impact PET radiopharmaceutical production by providing frameworks for waste management and sustainability. For example, radionuclide transport models [92] are adapted to evaluate effluent risks from short-lived isotopes like ¹⁸F-FDG, quantifying leakage probabilities and minimizing groundwater contamination during disposal [90]. These methodologies leverage advanced techniques, including multi-scale streamline approaches for real-time monitoring of radiation leaks via IoT sensors, as demonstrated in recent studies on eco-friendly decontamination using graphene oxide or solid-phase materials [93]. This enhances environmental risk components in integrated indicators, promoting energy-efficient processes and regulatory compliance under GMP standards, while differing from occupational-focused disciplines by prioritizing ecological modeling over human factors.

2.

Occupational Health and Safety: References [59,62,67,71,72] explore workplace risks, such as accidents, injuries, and health hazards, emphasizing the integration of subjective perceptions (e.g., worker surveys) with technical safety assessments. These studies commonly employ methodologies like fuzzy FMEA, risk matrices, and expert evaluations to assess and prioritize risks. For example, Ref. [59] utilizes KEMIRA-M and DEMATEL to enhance occupational safety in automobile glass manufacturing, while Ref. [67] applies a hybrid FMEA-FAHP-FTOPSIS approach for aviation maintenance, showcasing tailored strategies to improve workplace safety through combined subjective and technical insights.

3.

Civil Engineering/Construction: Research in Refs. [19,38,58,64] explores risks in infrastructure projects, focusing on seismic resilience, construction safety, and sustainability. These studies integrate technical, environmental, and occupational risks, utilizing advanced methodologies, such as performance-based engineering (PBE), Monte Carlo simulations, and machine learning. For instance, Ref. [19] employs PBE and fragility curves to evaluate seismic risks, while Ref. [38] develops digital safety risk models for highway construction, highlighting innovative approaches to improve infrastructure outcomes.

4.

Mining Engineering: References [45,49,53,82] investigate safety and environmental risks in mining operations, blending technical assessments with occupational and environmental considerations. These studies leverage advanced methodologies, including FMEA, system dynamics, and fuzzy set theory, to enhance risk management. For instance, Ref. [53] applies FMEA to identify hazards in coal mines, while Ref. [49] utilizes system dynamics for safety risk management, offering practical frameworks to improve mining safety and sustainability.

5.

Maritime Safety: References [65,84,86] address risks in maritime contexts, including navigation hazards, environmental conditions (e.g., weather and sea states), and occupational safety (e.g., mariner skills). Reference [86] employs adaptive multi-source risk quantification, a spatiotemporal modeling approach, to dynamically assess navigation risks. Reference [65] uses group FMEA to identify and prioritize failure modes in LNG carriers, focusing on operational safety. Reference [84] leverages AI-enhanced models to manage environmental and occupational risks. These studies collectively showcase diverse methodologies—spatiotemporal modeling, group FMEA, and AI—to improve maritime safety.

6.

Financial/Economic Risk: References [39,41] integrate economic, environmental, and geopolitical risks, focusing on their macroeconomic impacts and asset prices. They employ advanced econometric models, including VAR (vector autoregression), Diebold-Yilmaz FEVD (forecast error variance decomposition), and robust stochastic programming. Notably, Ref. [39] uses VAR and Fama-MacBeth regressions to assess global risks, emphasizing a comprehensive approach to understanding risk interactions and their economic effects.

7.

Systems Engineering: References [40,60,81] evaluate risks across political, financial, and environmental dimensions using system dynamics and integrated frameworks. Their methodological approaches include Ref. [40] employing system dynamics to model risk entropy, Ref. [81] utilizing G1 and entropy methods to assess risks in wind energy projects, and Ref. [60] applying Bayesian networks to integrate multidimensional risk factors. Together, these studies showcase advanced analytical tools for comprehensive risk management.

The methodologies differ based on the components integrated, the specific methods used, and the data sources, as summarized below:

• Integration of Components:

  ○

  Environmental science/engineering studies often integrate technical and environmental components, with less emphasis on subjective or occupational risks (e.g., Ref. [79] focuses on salinization and heavy metal contamination without subjective perception).

  ○

  Occupational health and safety studies integrate subjective perceptions and occupational risks, with varying environmental focus (e.g., Ref. [59] includes workplace hazards but focuses on occupational safety).

  ○

  Civil engineering integrates technical, environmental, and sometimes occupational risks, depending on the project scope (e.g., Ref. [19] includes CO₂ emissions and seismic resilience).

  ○

  Mining engineering integrates technical, environmental, and occupational risks, often using system dynamics for holistic assessment (e.g., Ref. [49] assesses subsystem impacts).

  ○

  Maritime safety integrates environmental, occupational, and technical risks, focusing on navigation and safety (e.g., Ref. [86] uses multi-source fusion for risk quantification).

  ○

  Financial/economic risk integrates economic and environmental factors, with less occupational focus (e.g., Ref. [39] assesses WEF risks without occupational components).

  ○

  Systems engineering integrates multiple dimensions, often using system dynamics to synthesize political, financial, and environmental risks (e.g., Ref. [40] combines risk entropy across dimensions).

• Methodological Approaches:

  ○

  Fuzzy logic and FMEA: Common in occupational safety and engineering (e.g., Ref. [37] uses fuzzy FMEA for wastewater treatment, and Ref. [28] for autonomous maintenance).

  ○

  Bayesian networks: Used in environmental and systems engineering (e.g., Ref. [27] for Na-Tech risks).

  ○

  Monte Carlo simulations: Prevalent in environmental and civil engineering (e.g., Ref. [31] for urban river systems, and Ref. [19] for seismic risk).

  ○

  System dynamics: Used in mining and systems engineering (e.g., Ref. [49] for coal mine safety, and Ref. [81] for wind energy projects).

  ○

  Multi-criteria decision analysis (MCDA): Applied across disciplines (e.g., Ref. [77] uses DEMATEL and AHP for environmental management, and Ref. [65] for maritime safety).

  ○

  Machine learning and AI: Emerging in environmental risk management and construction safety (e.g., Ref. [54] uses DFNN for hurricane response, and Ref. [38] for highway safety).

• Risk Indicators:

  ○

  Integrated risk indicators vary from simple indices like Risk Priority Number (RPN) in FMEA (e.g., [53]) to complex composite scores like systemic risk indicators (e.g., [43]) or resilience indices (e.g., [19]).

  ○

  Environmental studies may use risk quotients (e.g., [61]) or GeoPolRisk (e.g., [44]), while occupational studies use closeness coefficient indices (e.g., [67]).

• Data Sources:

  ○

  Subjective data (e.g., expert judgments and surveys) are common in occupational health (e.g., Ref. [71] uses worker surveys) and systems engineering (e.g., Ref. [81] uses expert scoring).

  ○

  Objective data (e.g., simulations and historical records) is prevalent in environmental science (e.g., Ref. [31] uses mechanistic models) and civil engineering (e.g., Ref. [20] uses probabilistic life-cycle analysis).

(Q2) Which risk assessment methods (e.g., probabilistic modeling, machine learning, or multi-criteria decision-making) are most used to evaluate risks and what are their strengths and limitations?

The review identified probabilistic modeling, machine learning, MCDM, and FMEA as the most prevalent methods for PET risk assessment, each with distinct strengths and limitations, supported by the Methods Table and external sources:

• Probabilistic modeling: Includes Monte Carlo simulations and fault tree analysis (FTA). Reference [75] uses Monte Carlo for radionuclide transport, offering robust uncertainty quantification but requiring extensive data, as noted in [94]. Strengths include handling complex systems, while limitations include computational complexity and data dependency, as seen in [84].
• Machine learning: Encompasses neural networks, random forests, and support vector machines. Reference [63] uses BP neural networks for safety prediction, achieving 80–90% accuracy, excelling in pattern recognition, as supported by [66]. Strengths include real-time adaptability, while limitations include opacity and data quality needs, as noted in [40].
• Multi-criteria decision-making (MCDM): Includes AHP, TOPSIS, and DEMATEL. Reference [62] uses fuzzy TOPSIS for risk prioritization, integrating criteria like probability and outcome, as supported by [80]. Strengths include handling subjectivity, while limitations include potential bias in expert weights, as in [76].
• Failure mode and effects analysis (FMEA): Widely used for quality risk management. Reference [95] applies FMEA to PET production, identifying critical phases like dispensing, aligning with FDA GMP. Strengths include systematic risk identification, while limitations include subjectivity and limited dynamic assessment, as in [72].

These methods balance technical rigor and practical applicability, with prevalence varying by focus (e.g., machine learning for prediction and FMEA for quality control), as evidenced by the Methods Table 1.

(Q3) How can existing integrated risk indicator frameworks from other industries (e.g., environmental science, nuclear safety, or maritime engineering) be adapted to address the specific occupational, technological, and environmental risks in PET radiopharmaceutical production?

Building on the core PET-specific risks outlined in the Introduction Section (e.g., short isotope half-lives and closed production systems), frameworks from other industries can be adapted as follows:

• Nuclear safety frameworks: Methods like AI-enhanced probabilistic risk evaluation [78] can be tailored to model time-critical failures in PET production, such as cyclotron downtime during the narrow ¹⁸F decay window, predicting occupational exposure spikes and integrating real-time dosimetry data. Limitations include adapting these to closed systems, where sensor placement must avoid compromising sterility—addressed by incorporating IoT for non-invasive monitoring [96].
• Environmental science approaches: Radionuclide transport models [69] from groundwater assessments can be adapted to evaluate PET waste management risks, quantifying leakage probabilities in closed effluent systems and integrating half-life decay kinetics for dynamic environmental impact scoring. This enhances pertinence by focusing on short-lived isotopes, unlike longer-lived nuclear waste, ensuring frameworks account for rapid decay in risk prioritization.
• Chemical processing techniques: FMEA and HAZOP from petrochemicals [20,21] can prioritize failure modes in PET’s closed hot cells, such as ventilation failures leading to radiation buildup. Adaptations involve weighting factors for half-life urgency (e.g., time-to-failure thresholds) and occupational components (e.g., finger dosimetry for manual interventions), making the analysis more relevant than generic applications.

(Q4) How do current studies quantify and integrate diverse risk factors (e.g., radiation exposure, equipment failure, and waste management) into a unified risk indicator for PET radiopharmaceutical production?

Current studies integrate diverse risk factors (radiation exposure, equipment failure, and waste management) into unified risk indicators using methodologies evidenced by the Methods Table and external sources:

• Weighted scoring: Reference [62] uses fuzzy TOPSIS to weight criteria (e.g., risk probability: 0.174), producing a closeness coefficient that integrates radiation exposure, equipment failure, and environmental risks, as supported by [95].
• Probabilistic methods: Reference [75] employs Monte Carlo simulations for radionuclide transport, integrating release dose, breakthrough time, and peak dose, addressing environmental and technological risks, as aligned with [94].
• Fuzzy logic: Reference [35] uses catastrophe progression for individual risk (IR) values, integrating pipeline risks adaptable to PET’s waste management, handling uncertainty, as seen in [40].
• Failure mode and effects analysis (FMEA): Reference [95] applies FMEA to identify critical PET production phases (e.g., dispensing), integrating occupational and technological risks via RPN, as supported by [72].
• Multi-criteria decision analysis (MCDA): Reference [80] uses fuzzy AHP-TOPSIS for overall performance score (OPS), integrating safety, environmental, and economic risks, as noted in [76].

These methods synthesize multiple factors, ensuring holistic assessment, though challenges include data variability and integration complexity, as highlighted in [66] and the FDA Briefing Document.

(Q5) What are the gaps in the literature regarding the application of integrated risk indicators in PET radiopharmaceutical production, and what future research directions can address these deficiencies?

• Limited stakeholder engagement: Few studies, like [34], emphasize stakeholder-driven co-design, limiting integration of risk perceptions, as noted in [83]. PET-specific stakeholder engagement, including operators and regulators, is underexplored.
• Lack of standardized frameworks: Methodological variability (e.g., PRA vs. FMEA) indicates a need for PET-specific standards, as highlighted in [66] and ISO 31000, noting inconsistent indicator definitions.
• Insufficient real-time monitoring: Limited use of real-time data, with potential for IoT and AI, as in [63], is under-represented in PET literature.
• Under-represented environmental risks: Waste management and emissions risks are under-addressed, with few studies focusing on PET-specific ecological impacts.

Future research directions include:

• PET-specific tools: Develop standardized risk assessment tools integrating technical, human, and environmental factors, as suggested by [24].
• AI and IoT integration: Enhance real-time monitoring and predictive modeling, aligning with [63,84].
• Standardized frameworks: Establish PET-specific risk indicator standards, as proposed in [66], to ensure consistency and benchmarking.
• Environmental risk assessment: Expand studies on radioactive waste and emissions, aligning with [75,95].

The systematic review of 70 references highlights those most applicable to PET radiopharmaceutical production risk assessment, focusing on methodologies like FMEA, risk perception, life-cycle, environmental PSR, and occupational risk. Key studies include Abuhussain in 2024 [62], utilizing fuzzy TOPSIS with risk perception surveys for green construction, Alizadeh [37], applying fuzzy FMEA to wastewater treatment for technological and environmental risks, Padgett [20], emphasizing life-cycle sustainability for long-term risk management, and Jing [82], integrating environmental and occupational risks using HFACS and Apriori algorithms. These references align with PET’s need for integrated risk indicators, addressing radiation exposure, equipment failure, and waste management. Their approaches, such as FMEA and stakeholder perception, ensure robust risk assessment, adaptable to PET’s safety and regulatory demands, enhancing occupational and environmental protection.

4. Conclusions

This systematic review of 70 studies from 2020 to 2025 provides a detailed analysis of risk assessment methodologies for PET radiopharmaceutical production, focusing on occupational, technological, and environmental risks. Predominant methods include probabilistic modeling, machine learning, multi-criteria decision-making, and failure mode and effects analysis, which effectively integrate complex risk factors, such as radiation exposure, equipment failure, and waste management, into unified indicators. Probabilistic modeling excels in quantifying uncertainties, while machine learning offers high predictive accuracy for real-time applications, achieving up to 80–90% precision in safety predictions. Multi-criteria decision-making and failure mode analysis ensure systematic hazard identification and prioritization, though they are constrained by data dependency and expert subjectivity. These methodologies align with regulatory standards like Good Manufacturing Practices, ensuring robust risk management in high-stakes environments. The interdisciplinary approach, spanning environmental science, occupational health, and engineering, underscores the need for comprehensive frameworks tailored to PET production’s unique challenges.

This systematic review faced several limitations: search and selection biases from English-only databases, potentially missing gray or non-English literature, high heterogeneity across interdisciplinary studies (e.g., civil engineering and mining), preventing meta-analysis and introducing subjectivity in adaptations, under-representation of direct PET-focused research, relying on analogous fields, temporal constraints excluding pre-2020 foundational works, absence of formal quality appraisals or cost analyses due to resources, and challenges in data extraction from truncated sources. Generalizability remains theoretical without primary PET data, though mitigated by PRISMA adherence and reference snowballing. Future work should expand scopes for robustness.

The methodological limitations of this systematic review stem from its aggregated synthesis of 70 studies (2020–2025), yielding an “average” risk profile without granular institution-specific comparisons, constrained by literature heterogeneity in facility scale, age, and resources, which hinders pinpointing superior practices. Approaches to identify best practices include adapting interdisciplinary frameworks—e.g., nuclear safety’s FMEA for occupational hazards and environmental science’s life-cycle assessments for waste impacts—coupled with longitudinal benchmarking using metrics like automation efficacy, dosimetry thresholds, and CO₂ emissions to evaluate and propagate optimal, GMP-compliant strategies in PET radiopharmaceutical production.

Despite these advancements, critical gaps persist. Limited stakeholder engagement hinders the incorporation of diverse risk perceptions, while inconsistent frameworks lack PET-specific standardization. Real-time monitoring remains underdeveloped, restricting dynamic risk assessment capabilities. Environmental risks, particularly radioactive waste and emissions, are underexplored, posing sustainability challenges. Future research should prioritize developing standardized, PET-specific risk tools that leverage AI and IoT for real-time monitoring. Establishing consistent methodologies will enhance benchmarking, and expanding environmental assessments will address waste management concerns. These targeted improvements will strengthen safety, sustainability, and regulatory compliance, ensuring effective risk management in PET radiopharmaceutical production’s complex, high-risk context.

This review establishes a foundational baseline for evaluating integrated risk indicators in PET radiopharmaceutical production, synthesizing methodologies and applications across occupational, technological, and environmental dimensions. Future investigations should prioritize comparative benchmarking of high-performing facilities, utilizing metrics such as automation efficacy, real-time monitoring integration, and CO₂ emission profiles, to delineate and propagate optimal practices tailored to institutional variability.